Options-driven Volatility Forecasting

We augment the Heterogeneous Autoregressive Regression model for forecasting realized volatility, using various measurements for the daily, weekly, and monthly volatilities, in addition to other predictive features. The main focus is on novel methods for extracting volatility estimators using option price data. Firstly, we provide a dimensionality reduction method for implied volatility surfaces built under the Black-Scholes model, whereby we combine simple row-wise and column-wise decompositions of the implied volatility surface with principal component analysis. Secondly, we provide a method for extracting the implied volatility under the Heston and Bates models. This is achieved by a calibration of these models while assuming that some of the model parameters remain constant. We demonstrate that these augmentations result in improved daily forecasts for realized volatility in a selection of different stocks. These volatility forecasts can also be utilized to further increase predictive performance for the realized volatility of other instruments, and can be combined for forecasting VIX.